Teamwork Graph Benchmark

Find out whether Teamwork Graph helps your AI agent give more useful answers about your real work.

The benchmark asks your agent the same question twice. The first answer uses only the agent's standard Atlassian access. The second answer uses Teamwork Graph. You then read both answers without being told which setup produced each one and choose the one you would rather use. The benchmark turns those choices into a report that compares answer quality, token use, and time.

Everything runs on your own computer against your own Atlassian site. Nothing is sent to Atlassian unless you decide to package a report and share it.

Why run it

• Decide whether Teamwork Graph improves answers for the work your team actually does.
• See which kinds of tasks benefit most from connected work context.
• Check whether token savings arrive without a drop in answer quality.
• Produce evidence you can use when deciding where to adopt or expand Teamwork Graph.

What the benchmark compares

The benchmark keeps the agent, the question, and the model fixed, so Teamwork Graph is the only meaningful difference between the two answers. Each side is called a context arm.

You see these two names throughout the benchmark, from setup to the final report. In commands, the same two arms are named control and test, and those literal names appear only when you run a benchmark from the command line.

In the review step the two answers appear as Answer A and Answer B, assigned so that you cannot tell which arm produced which answer. That is what makes your choice usable as evidence.

How it works

You install two command line tools and run one command. That command opens the benchmark app, a local web app in your browser that carries you through the whole benchmark. It is not a status dashboard you watch from the side: setup, prompt selection, the run, your review, and the final report all happen in it.

1. Install and set up. Install the Teamwork Graph CLI and the Benchmark CLI, then open the benchmark app and let it check your agent, sign-ins, and data sources.
2. Run a benchmark. Choose what to test, confirm the expected usage, and start the run. Both context arms answer the same prompts.
3. Review the answers. Read the two answers for each prompt and choose the more useful one. The app hides which arm produced each answer.
4. Read the report and share it. The app builds a report from your choices, token use, and timings. Sharing it with Atlassian is optional.

Expect 20 to 40 minutes end to end for a first run, most of it unattended while the two arms answer your prompts. Install and set up opens with everything you need before you start.

Get started

Install and set up the benchmark